VERIFY: A Framework for Physician Oversight of Ambient AI Scribe–Generated Clinical Documentation (Preprint)

<sec> <title>UNSTRUCTURED</title> <p>Background: Ambient AI scribes are deployed in approximately 30% of US health systems, with adoption doubling annually. Existing governance frameworks address organizational, system, regulatory, and medico-legal levels, but none equip individual physicians to evaluate AI-generated documentation before signing it into the permanent record. Objective: To propose a physician-level framework grounded in published evidence of AI scribe failure modes, including 1–3% hallucination, ~3% omission, fabricated medications, invented exam findings, incorrect negations, and misattributed statements. Discussion: The VERIFY framework is an integrated toolkit comprising a six-item mnemonic review protocol performed at the moment of signature, a four-tier risk taxonomy organized by harm potential, and pre-adoption vendor questions. The framework operationalizes the most consequential step in AI-assisted documentation: the moment between AI draft and signed medical record. Conclusion: VERIFY functions as the physician-level complement to existing organizational, regulatory, and tool-evaluation frameworks, completing governance at the point of care.</p> </sec>